Use this URL to cite or link to this record in EThOS:
Title: Three-dimensional geometry inference of convex and non-convex rooms using spatial room impulse responses
Author: Lovedee-Turner, Michael
ISNI:       0000 0004 8510 8482
Awarding Body: University of York
Current Institution: University of York
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
This thesis presents research focused on the problem of geometry inference for both convex- and non-convex-shaped rooms, through the analysis of spatial room impulse responses. Current geometry inference methods are only applicable to convex-shaped rooms, requiring between 6--78 discretely spaced measurement positions, and are only accurate under certain conditions, such as a first-order reflection for each boundary being identifiable across all, or some subset of, these measurements. This thesis proposes that by using compact microphone arrays capable of capturing spatiotemporal information, boundary locations, and hence room shape for both convex and non-convex cases, can be inferred, using only a sufficient number of measurement positions to ensure each boundary has a first-order reflection attributable to, and identifiable in, at least one measurement. To support this, three research areas are explored. Firstly, the accuracy of direction-of-arrival estimation for reflections in binaural room impulse responses is explored, using a state-of-the-art methodology based on binaural model fronted neural networks. This establishes whether a two-microphone array can produce accurate enough direction-of-arrival estimates for geometry inference. Secondly, a spherical microphone array based spatiotemporal decomposition workflow for analysing reflections in room impulse responses is explored. This establishes that simultaneously arriving reflections can be individually detected, relaxing constraints on measurement positions. Finally, a geometry inference method applicable to both convex and more complex non-convex shaped rooms is proposed. Therefore, this research expands the possible scenarios in which geometry inference can be successfully applied at a level of accuracy comparable to existing work, through the use of commonly used compact microphone arrays. Based on these results, future improvements to this approach are presented and discussed in detail.
Supervisor: Murphy, Damian Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available